Improving Robustness in Knowledge Distillation Using Domain-Targeted Data Augmentation